正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Before Fable got released (and pulled) Mozilla was quietly testing Claude Mythos against Firefox's 10M line codebase. The result? Over 400 security bugs fixes, including ones that had been hiding in the codebase for over a decade. Brian Grinstead, distinguished engineer at Mozilla, walked me through the agentic bug-finding... show more

claire vo 🖤

68,505 subscribers

260,557 次观看 • 4 个月前 •via X (Twitter)

教育健康养生科学技术

Anya Rossi• Live Now

Private livecam show

31 条评论

claire vo 🖤 的头像

claire vo 🖤4 个月前

Ty ty to our amazing sponsors: 💼 @WorkOS - Make your app enterprise ready today 🏆 @MetaviewAI - the agentic recruiting platform for winning teams 🎤 Listen everywhere: ▶️ YouTube: 🎧 Spotify: 🍎 Apple:

claire vo 🖤 的头像

claire vo 🖤4 个月前

Full breakdown on the @chatprd blog:

Morgan 的头像

Morgan4 个月前

@hnshah @mozilla @bgrins Watching this for Brian’s loop pattern ideas!

Sajid AI 的头像

Sajid AI4 个月前

@mozilla @bgrins Insane reminder that most “AI breakthroughs” are really just better agent loops + verification, not magic models.

Fernando Moreira 的头像

Fernando Moreira4 个月前

This is exactly why deterministic enforcement matters more than ever. Even when agents find real value (like the 400+ bugs at Mozilla), the risk of them making destructive changes in production codebases is real — especially with frontend stacks using Cursor/Claude. I built Nemesis Defender for this: Pretool Hook (exit(2) before execution) + Iron Dome scanner + eBPF kernel protection. 100% local, tamper-proof, calibrated for Next.js/React/TS. Project + demo: Would love your thoughts on agent harnesses vs hard boundaries, Claire.

Rob | Magnanimous Craft | Custom Waves @ 1Heal.org 的头像

Rob | Magnanimous Craft | Custom Waves @ 1Heal.org4 个月前

@mozilla @bgrins friend developed a faster code search tool to make this easier called src-nav which does things grep cannot

Brooke Lacey 的头像

Brooke Lacey4 个月前

@mozilla @bgrins Ma’am I know I probs come off as a weird fangirl but these episodes are so informative and helpful - thank you for doing what you do! 😍

Sebastian Buzdugan 的头像

Sebastian Buzdugan4 个月前

@mozilla @bgrins finding 400 bugs is easier than proving the patches don't add regressions

Rafael Pinheiro 的头像

Rafael Pinheiro4 个月前

@mozilla @bgrins To point a model deterministically where duplications are in large codebases, dupehound does this really well, superfast and open-source.

Brosko 的头像

Brosko4 个月前

@mozilla @bgrins the 'you can build this in an afternoon' part is either the most motivating or most humbling thing ive read all week no in between tbh 😭

Abdulmuiz Adeyemo 的头像

Abdulmuiz Adeyemo4 个月前

@mozilla @bgrins Insane

AI Mastery Guide 的头像

AI Mastery Guide4 个月前

@mozilla @bgrins What's the "lie to the agent" part about, curious how that helped with false positives

Hidden Tape 的头像

Hidden Tape4 个月前

@mozilla @bgrins 400 fixes sounds great but id wanna know how many of those got properly reviewed before merge agent found it doesnt mean a human understood it

Sai Charan L 的头像

Sai Charan L4 个月前

@mozilla @bgrins The "50% setup" line is the whole lesson. Everyone argues about which model. The model is the cheap part now. The harness is the skill: a tight goal/loop and a verifier that kills false positives. Same model, two setups, completely different output.

Ralf Kronen 的头像

Ralf Kronen4 个月前

@mozilla @bgrins The "afternoon to write your own harness" line is the headline. If that harness plus a verifier yields 400 confirmed bugs, the result is a property of the setup, not of Mythos. Any capable model fits that slot. The scarce part was never the model, it was the scaffolding.

el_cuda 的头像

el_cuda4 个月前

@mozilla @bgrins I wanna know what lies to tell!

晚晚的头像

晚晚4 个月前

@mozilla @bgrins 十年老 bug 被翻出来，这就有点狠了

Yossi Eliaz 的头像

Yossi Eliaz4 个月前

@mozilla @bgrins Yes, agent power stems from the model and harness, like Schumacher and a Ferrari.

Jens Honack 的头像

Jens Honack4 个月前

@mozilla @bgrins Mythos was able to find some really gnarly bugs too. Impressive considering Firefox has some extremely bright engineers. As competent the human mind is, it's simply not as good a repository of structured information, as LLMs, which is probably why LLMs excel at this.

Simply Complex Systems 的头像

Simply Complex Systems4 个月前

@mozilla @bgrins The only gates to a deeply integrated, self directed agentic environment are careful reading and well articulated questions.

Picassio 的头像

Picassio4 个月前

@mozilla @bgrins lying to the agent to get better results is the part im stuck on that sounds wild and i need the whole explanation NOW

Gregor 的头像

Gregor4 个月前

@mozilla @bgrins not sure 400 fixes is the headline here, curious what the validation loop actually looked like. at 10M lines, how many of those patches had a human review before merge?

Pawzard 的头像

Pawzard4 个月前

@mozilla @bgrins decade old bugs just vibing in the codebase like a tenant who stopped paying rent in 2014 but nobody checked 💀

★ Adam LG Ring ★ BSc - IBM 的头像

★ Adam LG Ring ★ BSc - IBM4 个月前

@mozilla @bgrins the load on newly released models is going to be insane from now on who knows how long it will be up !!!

A 的头像

A4 个月前

@mozilla @bgrins the model isn't the product. the loop around the model is where the leverage hides.

David Jones 的头像

David Jones4 个月前

In the agentic security bug pipeline for Firefox, how did you handle prioritizing and sequencing work across the codebase (or potential issues), respecting compute costs, dependencies, or human review bandwidth — and did the harness integrate directly with existing bug tracking/backlog tools, or was that kept fully human-driven?

Axel Winter 的头像

Axel Winter4 个月前

@mozilla @bgrins 499 isnt much though

Benign Rad-Science 的头像

Benign Rad-Science4 个月前

@mozilla @bgrins Don’t you think it’s wierd that Anthropic have it to Mozilla for a browser rather than to the US govern to secure its systems? Wierd.

Prakhar 的头像

Prakhar4 个月前

@mozilla @bgrins I feel this is like tech debt and you have to get to it sooner or later. But agents make it possible to do so fast and well, that it is potentially not as big of a deal as before. I see companies releasing their stacks in rust so frequently now.

Adel Bucetta 的头像

Adel Bucetta4 个月前

@mozilla @bgrins this is massive. the real unlock isn't just testing your model against a big codebase, it's what that says about the state of code maintenance and how we need tools like claude to catch up with our own sloppiness

Makaroni 的头像

Makaroni4 个月前

@mozilla @bgrins the 50/50 split is the real takeaway people keep crediting the model but the harness doing half the work is not a small detail

相关视频

Claude Code Unpacked! Visual walkthrough of the entire 500k-line leaked codebase. What happens when you type a message: - the agent loop - 50+ tools - multi-agent orchestration - unreleased features Want to understand the internals or build your own agent harness? start here:

Claude Code Unpacked! Visual walkthrough of the entire 500k-line leaked codebase. What happens when you type a message: - the agent loop - 50+ tools - multi-agent orchestration - unreleased features Want to understand the internals or build your own agent harness? start here:

Akshay 🚀

24,552 次观看 • 6 个月前

The founder of LangChain says both models and harnesses have gotten really good between December and now. According to Harrison Chase, the core idea of an agent before Christmas was a model running in a loop and calling tools. This had been the north star for 3 years. - langchain had this when it launched - autogpt was the same idea - openclaw is kind of a future version of it Then about a year ago, they started getting really good. Claude Code, Manus, and Deep Research were all launched around the same time. All of them use the same pattern: running in a loop with harnesses (planning tools, file systems, code execution, etc) Harness engineering became a thing. Then Opus came out in November and really unlocked it. - the harness let the model do more and more - less hardcoded logic - way more control Then everyone went on vacation, played around, and realized that the model and the harness finally worked reliably.

The founder of LangChain says both models and harnesses have gotten really good between December and now. According to Harrison Chase, the core idea of an agent before Christmas was a model running in a loop and calling tools. This had been the north star for 3 years. - langchain had this when it launched - autogpt was the same idea - openclaw is kind of a future version of it Then about a year ago, they started getting really good. Claude Code, Manus, and Deep Research were all launched around the same time. All of them use the same pattern: running in a loop with harnesses (planning tools, file systems, code execution, etc) Harness engineering became a thing. Then Opus came out in November and really unlocked it. - the harness let the model do more and more - less hardcoded logic - way more control Then everyone went on vacation, played around, and realized that the model and the harness finally worked reliably.

Ivan Burazin

33,948 次观看 • 6 个月前

Claude Code is the OpenClaw alternative you already had. I've been telling people this in person for three months. Now it's a piece on Every 📧. When OpenClaw was at peak hype a few months ago, everyone was going crazy over it — for good reasons. But almost nobody made the obvious comparison: Claude Code already does the thing you love about OpenClaw. In a lot of cases, better. Here's why people missed it. You have to understand the difference between a model and a harness. OpenClaw, at the end of the day, is just a harness on top of AI models. And Claude Code is too — it just got marketed as a coding tool, an alternative to Cursor. Until Cowork launched, almost nobody looked at it as a tool for non-programmers. Dan Shipper 📧 was one of the earliest to make that claim, and he was damn right. Let Claude Code run from the home folder of your computer with free access, and you get exactly what made OpenClaw exciting. Instead of just arguing it, here's a video walkthrough of a couple of my AI employees i built on the Claude Code harness. And here's the full article:

Claude Code is the OpenClaw alternative you already had. I've been telling people this in person for three months. Now it's a piece on Every 📧. When OpenClaw was at peak hype a few months ago, everyone was going crazy over it — for good reasons. But almost nobody made the obvious comparison: Claude Code already does the thing you love about OpenClaw. In a lot of cases, better. Here's why people missed it. You have to understand the difference between a model and a harness. OpenClaw, at the end of the day, is just a harness on top of AI models. And Claude Code is too — it just got marketed as a coding tool, an alternative to Cursor. Until Cowork launched, almost nobody looked at it as a tool for non-programmers. Dan Shipper 📧 was one of the earliest to make that claim, and he was damn right. Let Claude Code run from the home folder of your computer with free access, and you get exactly what made OpenClaw exciting. Instead of just arguing it, here's a video walkthrough of a couple of my AI employees i built on the Claude Code harness. And here's the full article:

Nityesh

22,895 次观看 • 4 个月前

Andrew Ng just dropped a free course on Claude Code from scratch, taught with the Anthropic team: 00:00 - why Claude Code is so agentic 04:00 - shockingly simple architecture 12:00 - point it at any codebase this short watch will replace 10 paid coding agent courses. Andrew Ng calls it his personal favorite coding assistant right now. Watch it today, then read how to engineer your own agent loops in the article below

Andrew Ng just dropped a free course on Claude Code from scratch, taught with the Anthropic team: 00:00 - why Claude Code is so agentic 04:00 - shockingly simple architecture 12:00 - point it at any codebase this short watch will replace 10 paid coding agent courses. Andrew Ng calls it his personal favorite coding assistant right now. Watch it today, then read how to engineer your own agent loops in the article below

Carnage

715,565 次观看 • 3 个月前

This is one of the coolest open-source AI agent projects I've seen in a while: 'Understand Anything' It's a plugin for Claude Code, Codex, OpenCode etc. that analyzes your codebase and turns it into a knowledge base that you can interact with. It explains the codebase to you, rather than showing you the structure. It seems like it's designed for code but I opened my Obsidian vault of podcast highlights in Claude Code, then ran /understand. The result is a knowledge graph that I can search of highlights from 888 podcast episodes and 144K lines of markdown text.

This is one of the coolest open-source AI agent projects I've seen in a while: 'Understand Anything' It's a plugin for Claude Code, Codex, OpenCode etc. that analyzes your codebase and turns it into a knowledge base that you can interact with. It explains the codebase to you, rather than showing you the structure. It seems like it's designed for code but I opened my Obsidian vault of podcast highlights in Claude Code, then ran /understand. The result is a knowledge graph that I can search of highlights from 888 podcast episodes and 144K lines of markdown text.

Dan McAteer

175,308 次观看 • 4 个月前

I'm sharing with you one of the best papers I've read on Harness Engineering it takes apart Claude Code (Opus 5.5), Codex, Gemini CLI and more, and boils all of them down to one line: agent = model + harness what you walk away with: > what a harness actually is > how today's coding agents are built under the hood > the patterns every one of them repeats > where the whole category is heading > their checklist for building your own the model gets the headlines. the harness decides whether the agent works loop and graph engineering on top of it - in my article below

I'm sharing with you one of the best papers I've read on Harness Engineering it takes apart Claude Code (Opus 5.5), Codex, Gemini CLI and more, and boils all of them down to one line: agent = model + harness what you walk away with: > what a harness actually is > how today's coding agents are built under the hood > the patterns every one of them repeats > where the whole category is heading > their checklist for building your own the model gets the headlines. the harness decides whether the agent works loop and graph engineering on top of it - in my article below

Mr. Buzzoni

84,231 次观看 • 23 天前

The easiest way to run your agent harnesses using local models: Whether you're using Claude Code, Codex, OpenCode, or Pi, the local setup itself isn't really the hard part. The harder problem is figuring out which local model your machine can actually handle well. You have to think about RAM, model size, quantization, context length, KV cache, speed, accuracy, and a bunch of other trade-offs before you even start. Magnitude is an open-source solution that removes all of that guesswork for you. It profiles your machine, benchmarks what it can realistically run, recommends the best models for your hardware, and then lets you connect them to your preferred agent harness. The entire setup takes just two commands. In this video, I walk through the full process from hardware profiling to running a harness on a local model. Chapters: 00:00 Intro 00:17 What Magnitude is 00:45 Setup in two commands 01:29 Why picking a local model gets confusing fast 02:11 Hardware profiling + model recommendations 02:59 Choosing your harness 03:22 Connecting a harness that's already running 04:18 Outro Get started: (don't forget to star 🌟)

The easiest way to run your agent harnesses using local models: Whether you're using Claude Code, Codex, OpenCode, or Pi, the local setup itself isn't really the hard part. The harder problem is figuring out which local model your machine can actually handle well. You have to think about RAM, model size, quantization, context length, KV cache, speed, accuracy, and a bunch of other trade-offs before you even start. Magnitude is an open-source solution that removes all of that guesswork for you. It profiles your machine, benchmarks what it can realistically run, recommends the best models for your hardware, and then lets you connect them to your preferred agent harness. The entire setup takes just two commands. In this video, I walk through the full process from hardware profiling to running a harness on a local model. Chapters: 00:00 Intro 00:17 What Magnitude is 00:45 Setup in two commands 01:29 Why picking a local model gets confusing fast 02:11 Hardware profiling + model recommendations 02:59 Choosing your harness 03:22 Connecting a harness that's already running 04:18 Outro Get started: (don't forget to star 🌟)

Akshay 🚀

18,722 次观看 • 1 个月前

A single Claude skill just passed 164,000 stars on GitHub. It's called Ponytail. And it makes your agent code like that senior dev who never writes a line more than needed. The one who looks at your 50 lines and replaces them with one. Version 5 just dropped: → 53% less code → 41% less time → 26% lower cost → 45% fewer tokens And the crazy part: It writes half the code, but doesn't skip the testing. 98% of critical code ships with a test to make sure it works. Without the skill, only 68%. Real example: You ask it to add a date picker. Without the skill → builds one from scratch in 335 lines With Ponytail → uses the browser's built-in one in 10 lines Why it matters to you: If you code with agents, you pay for every line they write. Less code = fewer tokens, fewer bugs and less to maintain. Works with Claude Code, Codex, Cursor, Copilot and 15+ other agents. 164K stars. MIT license. Free. Link below :)

A single Claude skill just passed 164,000 stars on GitHub. It's called Ponytail. And it makes your agent code like that senior dev who never writes a line more than needed. The one who looks at your 50 lines and replaces them with one. Version 5 just dropped: → 53% less code → 41% less time → 26% lower cost → 45% fewer tokens And the crazy part: It writes half the code, but doesn't skip the testing. 98% of critical code ships with a test to make sure it works. Without the skill, only 68%. Real example: You ask it to add a date picker. Without the skill → builds one from scratch in 335 lines With Ponytail → uses the browser's built-in one in 10 lines Why it matters to you: If you code with agents, you pay for every line they write. Less code = fewer tokens, fewer bugs and less to maintain. Works with Claude Code, Codex, Cursor, Copilot and 15+ other agents. 164K stars. MIT license. Free. Link below :)

Bruce Radley

244,198 次观看 • 12 天前

Entire CEO Thomas Dohmke on how his Codex agent ran for 42 hours over the weekend building the wrong thing entirely: "Over the weekend I had Codex write for 42 hours straight, working on something that I thought was a great plan." "On Monday it actually disconnected because my router restarted. I was like, where are we? And it's like, oh, I'm building this harness. And I'm like, wait a second, what harness are you building? My prompt said build a skill, not a harness." "For 42 hours the agent ran long and made me feel good while I was enjoying summer in Seattle. In reality it built something completely different than what my intent was." Entire

Entire CEO Thomas Dohmke on how his Codex agent ran for 42 hours over the weekend building the wrong thing entirely: "Over the weekend I had Codex write for 42 hours straight, working on something that I thought was a great plan." "On Monday it actually disconnected because my router restarted. I was like, where are we? And it's like, oh, I'm building this harness. And I'm like, wait a second, what harness are you building? My prompt said build a skill, not a harness." "For 42 hours the agent ran long and made me feel good while I was enjoying summer in Seattle. In reality it built something completely different than what my intent was." Entire

MTS

12,661 次观看 • 2 个月前

Run your coding agent on a local model, with QVAC. OpenCode now plugs into a model on your own machine through our official plugin. That changes two things. Privacy by default. Your repo, your prompts, and every command the agent runs stay on your device. No third-party model ever sees your code. For an agent that reads your codebase and runs commands, that is the point, not a nice-to-have. New workflows. With no meter and no rate limit, the agent does not have to stop. Let it loop on a task, refactor a whole repo overnight, or run a watch-build-fix loop that keeps going. Setup takes a few minutes: one line in opencode.json and OpenCode runs on a local QVAC model. In the video demo, it handles three real developer tasks back to back, fully offline.

Run your coding agent on a local model, with QVAC. OpenCode now plugs into a model on your own machine through our official plugin. That changes two things. Privacy by default. Your repo, your prompts, and every command the agent runs stay on your device. No third-party model ever sees your code. For an agent that reads your codebase and runs commands, that is the point, not a nice-to-have. New workflows. With no meter and no rate limit, the agent does not have to stop. Let it loop on a task, refactor a whole repo overnight, or run a watch-build-fix loop that keeps going. Setup takes a few minutes: one line in opencode.json and OpenCode runs on a local QVAC model. In the video demo, it handles three real developer tasks back to back, fully offline.

QVAC

11,382,936 次观看 • 4 个月前

I've long said that o3 is the best coding model - but if you're using an agent harness - Claude is just better at navigating your codebase. Enter the Repo Prompt pair programmer mode - it's the best of both words, as Claude coordinates with o3 to plan and apply edits for you!

I've long said that o3 is the best coding model - but if you're using an agent harness - Claude is just better at navigating your codebase. Enter the Repo Prompt pair programmer mode - it's the best of both words, as Claude coordinates with o3 to plan and apply edits for you!

eric provencher

93,571 次观看 • 1 年前

FREE CLAUDE FABLE 5 and almost nobody knows this trick Anthropic recently released their most capable model ever - the first Mythos-class model available to the public normally it's $10/$50 per million tokens but there's a hack: GitLab just added Fable 5 to Duo Agent Platform - across ALL tiers, including free trial the setup: 1/ register a GitLab account: 2/ start the free GitLab Duo trial 3/ Fable 5 is live through their AI Gateway 4/ done - Mythos-class model, $0 why Fable 5 is worth the hype: 1/ 80.3% SWE-Bench Pro - 11 points ahead of the next best model 2/ multi-DAY autonomous runs without losing the plot 3/ single-pass implementations of systems that took days of iteration 4/ Karpathy called it 'a major-version-bump-deserving step change' also free on Claude Pro/Max/Team until June 22 - after that it's usage credits you have a limited window to run the most capable public model on the planet for FREE

FREE CLAUDE FABLE 5 and almost nobody knows this trick Anthropic recently released their most capable model ever - the first Mythos-class model available to the public normally it's $10/$50 per million tokens but there's a hack: GitLab just added Fable 5 to Duo Agent Platform - across ALL tiers, including free trial the setup: 1/ register a GitLab account: 2/ start the free GitLab Duo trial 3/ Fable 5 is live through their AI Gateway 4/ done - Mythos-class model, $0 why Fable 5 is worth the hype: 1/ 80.3% SWE-Bench Pro - 11 points ahead of the next best model 2/ multi-DAY autonomous runs without losing the plot 3/ single-pass implementations of systems that took days of iteration 4/ Karpathy called it 'a major-version-bump-deserving step change' also free on Claude Pro/Max/Team until June 22 - after that it's usage credits you have a limited window to run the most capable public model on the planet for FREE

kaize

24,809 次观看 • 4 个月前

DeepSeek-V4.1-Flash: the BEST Boeing benchmark result I’ve seen from an open-source model (by far). ✈️ It ran for hours with /goal. Where other models often stall quickly, it kept improving: inspect, spot a defect, zoom in, diagnose, fix, repeat. The video really shows how strong it is at sustaining that process over a long horizon (and how good it was at identifying what to work on). Impressive. Harness: Claude Code Prompt: "/goal create the most realistic boeing 747 using THREEJS - use your vision capatilities to create a self verifiable system, enter a loop until you are 100% satified about the result (you can build a camera system to inspect each angle)"

DeepSeek-V4.1-Flash: the BEST Boeing benchmark result I’ve seen from an open-source model (by far). ✈️ It ran for hours with /goal. Where other models often stall quickly, it kept improving: inspect, spot a defect, zoom in, diagnose, fix, repeat. The video really shows how strong it is at sustaining that process over a long horizon (and how good it was at identifying what to work on). Impressive. Harness: Claude Code Prompt: "/goal create the most realistic boeing 747 using THREEJS - use your vision capatilities to create a self verifiable system, enter a loop until you are 100% satified about the result (you can build a camera system to inspect each angle)"

Victor M

71,631 次观看 • 1 个月前

Anthropic engineer: "The fastest way to make an agent better isn't a smarter model, it's the right loop and graph around it." In 60 minutes she breaks down how Anthropic engineers use Claude Code, build agents that catch their own mistakes and improve with every run. This beats any $500 agentic engineering course you can find. Watch it, then read the full guide on loops and graphs below.

Anthropic engineer: "The fastest way to make an agent better isn't a smarter model, it's the right loop and graph around it." In 60 minutes she breaks down how Anthropic engineers use Claude Code, build agents that catch their own mistakes and improve with every run. This beats any $500 agentic engineering course you can find. Watch it, then read the full guide on loops and graphs below.

Anatoli Kopadze

136,422 次观看 • 2 个月前

Google engineer: “Just delete your IDE—you don’t need it anymore. Ask Claude Code to run Claude Code for you. 85% of PRs at Google are already shipped by AI agents. If you aren’t building the harness around your agents, it’s crazy how far behind you are.” In this 90-minute talk, a Google engineer with 30 years of experience explains what the future of AI engineering will look like. Worth more than 10 paid agentic engineering courses. Watch it today, then read the article below to learn how to build a harness for reliable, self-improving AI agents.

Google engineer: “Just delete your IDE—you don’t need it anymore. Ask Claude Code to run Claude Code for you. 85% of PRs at Google are already shipped by AI agents. If you aren’t building the harness around your agents, it’s crazy how far behind you are.” In this 90-minute talk, a Google engineer with 30 years of experience explains what the future of AI engineering will look like. Worth more than 10 paid agentic engineering courses. Watch it today, then read the article below to learn how to build a harness for reliable, self-improving AI agents.

Ichigo

84,780 次观看 • 1 个月前

MIT GAVE DEBUGGING ITS OWN FULL LECTURE -- BECAUSE MOST DEVS STILL HUNT BUGS WITH PRINT STATEMENTS LIKE IT'S 1995 54 minutes that take the thing you do every single day, badly, and show you the tools schools never bothered to teach. -> The moment it lands, you realize "add a print, run it, add another print" was never debugging. It was guessing with extra steps. Most people never learned the real toolkit. A proper debugger, breakpoints, tracing what the program actually does instead of what you assume it does. In 2026 guessing isn't good enough -> the bug is buried in code you didn't even write, and printf won't find it for you. An AI agent will generate hundreds of lines in seconds. When one of them quietly breaks, finding it runs on exactly the skills in this lecture. Anyone can add a print statement. The person who can actually trace a bug to its source is the one who ships while everyone else guesses. Bookmark & Watch it ↓

MIT GAVE DEBUGGING ITS OWN FULL LECTURE -- BECAUSE MOST DEVS STILL HUNT BUGS WITH PRINT STATEMENTS LIKE IT'S 1995 54 minutes that take the thing you do every single day, badly, and show you the tools schools never bothered to teach. -> The moment it lands, you realize "add a print, run it, add another print" was never debugging. It was guessing with extra steps. Most people never learned the real toolkit. A proper debugger, breakpoints, tracing what the program actually does instead of what you assume it does. In 2026 guessing isn't good enough -> the bug is buried in code you didn't even write, and printf won't find it for you. An AI agent will generate hundreds of lines in seconds. When one of them quietly breaks, finding it runs on exactly the skills in this lecture. Anyone can add a print statement. The person who can actually trace a bug to its source is the one who ships while everyone else guesses. Bookmark & Watch it ↓

slash1s

10,864 次观看 • 4 个月前

this is pure f*cking gold for anyone running coding agents Jev founder Diogo Amogo wrote a PDF on building a Jev harness the promise: > 200x faster > 400x cheaper the model hasn't been the slow part for a while the speed and the cost sit in the harness wrapped around it • how to use it > drop this PDF and the article below into Claude Code or Codex > let it rebuild its own setup one evening of setup and your agent starts next week on a harness most teams still don't have 👇

this is pure f*cking gold for anyone running coding agents Jev founder Diogo Amogo wrote a PDF on building a Jev harness the promise: > 200x faster > 400x cheaper the model hasn't been the slow part for a while the speed and the cost sit in the harness wrapped around it • how to use it > drop this PDF and the article below into Claude Code or Codex > let it rebuild its own setup one evening of setup and your agent starts next week on a harness most teams still don't have 👇

Mr. Buzzoni

128,669 次观看 • 1 个月前

this is pure f*cking gold for anyone running coding agents Jev founder Diogo Amogo wrote a PDF on building a Jev harness the promise: > 200x faster > 400x cheaper the model hasn't been the slow part for a while the speed and the cost sit in the harness wrapped around it • how to use it > drop this PDF and the article below into Claude Code or Codex > let it rebuild its own setup one evening of setup and your agent starts next week on a harness most teams still don't have

this is pure f*cking gold for anyone running coding agents Jev founder Diogo Amogo wrote a PDF on building a Jev harness the promise: > 200x faster > 400x cheaper the model hasn't been the slow part for a while the speed and the cost sit in the harness wrapped around it • how to use it > drop this PDF and the article below into Claude Code or Codex > let it rebuild its own setup one evening of setup and your agent starts next week on a harness most teams still don't have

Mr. Buzzoni

78,793 次观看 • 11 天前

Karpathy said something you'll regret ignoring: "We have to keep the AI on the leash. I'm still the bottleneck. I have to make sure this thing isn't introducing bugs and that there's no security issues." He said it at YC talk last year, when the worry was reliability. The models hallucinated and made mistakes no human would, so the leash implied keeping yourself in the loop and checking the output before trusting it. The models are far better now, and the line still holds, for a reason he was not focused on back then. Even a model that writes flawless code today still has no idea who is allowed to run it. Correctness and authorization are different problems, and only correctness improves as the model improves. A perfect agent still hands a tool where anyone can do anything, because permission was never part of the task. I actually tested this in practice with Claude Code. I asked it to build a small internal tool with a button that issues account credits. It worked first try, and running it locally, the credit applied the instant I clicked. Nothing decided who was allowed to click it. The agent wrote the right logic and displayed a success notification. It never checked whether the caller had the right, whether it should pause for a human, or whether anything was logged. And this is not a bug a smarter model can outgrow because the leash was never in the code. Identity, permissions, and audit live in the system that runs the app, not in what the agent generates. To solve this, I took the exact same bundle and hosted it on Retool. The credit write that fired silently on my laptop now stopped at an approval gate, resolved to a real identity through SSO, and landed in an audit log. I wrote none of it. The app inherited the entire boundary the moment it was deployed, and the video shows the before and after. You can try it yourself here: I also wrote a detailed breakdown of the whole thing in my recent article, and I worked with the team to put this together. It walks through the build, the exact moment the credit write went through on my laptop with nobody checking, and then what changed when the same app ran on Retool. It also covers why this is a property of the runtime and not something a better model fixes, which is why devs typically miss this. The article is quoted below.

Karpathy said something you'll regret ignoring: "We have to keep the AI on the leash. I'm still the bottleneck. I have to make sure this thing isn't introducing bugs and that there's no security issues." He said it at YC talk last year, when the worry was reliability. The models hallucinated and made mistakes no human would, so the leash implied keeping yourself in the loop and checking the output before trusting it. The models are far better now, and the line still holds, for a reason he was not focused on back then. Even a model that writes flawless code today still has no idea who is allowed to run it. Correctness and authorization are different problems, and only correctness improves as the model improves. A perfect agent still hands a tool where anyone can do anything, because permission was never part of the task. I actually tested this in practice with Claude Code. I asked it to build a small internal tool with a button that issues account credits. It worked first try, and running it locally, the credit applied the instant I clicked. Nothing decided who was allowed to click it. The agent wrote the right logic and displayed a success notification. It never checked whether the caller had the right, whether it should pause for a human, or whether anything was logged. And this is not a bug a smarter model can outgrow because the leash was never in the code. Identity, permissions, and audit live in the system that runs the app, not in what the agent generates. To solve this, I took the exact same bundle and hosted it on Retool. The credit write that fired silently on my laptop now stopped at an approval gate, resolved to a real identity through SSO, and landed in an audit log. I wrote none of it. The app inherited the entire boundary the moment it was deployed, and the video shows the before and after. You can try it yourself here: I also wrote a detailed breakdown of the whole thing in my recent article, and I worked with the team to put this together. It walks through the build, the exact moment the credit write went through on my laptop with nobody checking, and then what changed when the same app ran on Retool. It also covers why this is a property of the runtime and not something a better model fixes, which is why devs typically miss this. The article is quoted below.

Akshay 🚀

42,911 次观看 • 4 个月前